Tilt-Based Automatic Zooming and Scaling in Mobile Devices – a State-Space Implementation
Total Page:16
File Type:pdf, Size:1020Kb
Tilt-Based Automatic Zooming and Scaling in Mobile Devices – a state-space implementation Parisa Eslambolchilar 1, Roderick Murray-Smith 1,2 1 Hamilton Institute, National University of Ireland, NUI, Maynooth, Co.Kildare , Ireland [email protected] 2 Department of Computing Science, Glasgow University, Glasgow G12 8QQ, Scotland [email protected] Abstract. We provide a dynamic systems interpretation of the coupling of in- ternal states involved in speed-dependent automatic zooming, and test our im- plementation on a text browser on a Pocket PC instrumented with an acceler- ometer. The dynamic systems approach to the design of such continuous interaction interfaces allows the incorporation of analytical tools and construc- tive techniques from manual and automatic control theory. We illustrate ex- perimental results of the use of the proposed coupled navigation and zooming interface with classical scroll and zoom alternatives. 1 Introduction Navigation techniques such as scrolling (or panning) and zooming are essential components of mobile device applications such as map browsing and reading text documents, allowing the user access to a larger information space than can be viewed on the small screen. Scrolling allows the user to move to different locations, while zooming allows the user to view a target at different scales. However, the restrictions in screen space on mobile devices make it difficult to browse a large document effi- ciently. Using the traditional scroll bar, the user must move back and forth between the document and the scroll bar, which can increase the effort required to use the in- terface. In addition, in a long document, a small movement of the handle can cause a sudden jump to a distant location, resulting in disorientation and frustration. Speed-dependent automatic zooming is a relatively new navigation technique [7, 8, 14, 22, 25, 26] that unifies rate-based scrolling and zooming to overcome these limita- tions. The user controls the scrolling speed only, and the system automatically adjusts the zoom level so that the speed of visual flow across the screen remains constant. Us- ing this technique, the user can smoothly locate a distant target in a large document without having to manually interweave zooming and scrolling, and without becoming disoriented by extreme visual flow. In this paper we demonstrate that, as suggested by Igarashi and Hinckley [14], SDAZ is well suited to implementation on mobile devices instrumented with tilt sen- sors, which can then be comfortably controlled in a single-handed fashion. We also describe an alternative stylus controlled implementation for the PocketPC. A further contribution is the use of a state-space formulation of speed dependent zooming, which we believe is a promising reformulation of the technique, which opens the path to the use of analytic tools from optimal and manual control theory. 2 Speed-dependent automatic zooming – a brief review Several techniques have been proposed to improve the manipulation of scroll bars [14, 19]. They allow the user to control scrolling speed, enabling fine positioning in large documents. LensBar [18] combines these techniques with interactive filtering and semantic zooming, and also provides explicit control of zooming via horizontal motion of the mouse cursor. A rate-based scrolling interface is described in [29] that maps displacement of the input device to the velocity of scrolling. Zoomable user interfaces, such as Pad and Pad++ [4], use continuous zooming as a central navigation tool. The objects are spatially organized in an infinite two- dimensional information space, and the user accesses a target object using panning and zooming operations. A notable problem with the original zoomable interfaces is that they require explicit control of both panning and zooming, and it is sometimes difficult for the user to coordinate them. The user can get lost in the infinite informa- tion space [16]. Bimanual approaches also exist, such as that of Guiard et al. [11] where a joystick in one hand controlled zoom level, and a mouse in the other provided navigation. They showed that by using zooming interfaces, bit rates far beyond those possible in physical selection tasks become possible. Information visualization techniques, such as Fisheye Views [9, 12], Perspective Wall [17], and the Document Lens [21] also address the problem of information over- load by distorting the view of documents. The focused area is magnified, while the non-focused areas are squashed but remain in spatial context. The user specifies the next focal point by clicking or panning. Van Wijk derived an optimal trajectory for panning and zooming in [24], for known start and end points. The particular input device used can also influence the effectiveness of rate con- trol. An experiment on 6 DOF input control [29] showed that rate control is more ef- fective with isometric or elastic devices, because of their self-centring nature. It is also reported that an isometric rate-control joystick [2] can surpass a traditional scroll bar and a mouse with a finger wheel [29]. Another possibility is to change the rate of scrolling or panning in response to tilt, as demonstrated by Rekimoto [20] as well as Harrison et al. [13], suitable for small screen devices like mobiles phones and PDAs. A common problem with scrolling and zooming interfaces is that when users are zoomed out for orientation, there is not enough detail to do any ‘real work’. When they are zoomed in sufficiently to see detail, the context is lost. To reduce this prob- lem, multiple windows can be provided, each with pan and zoom capability. Although this is reasonable for small information spaces, the many windows required by large spaces often lead to usability problems due to excessive screen clutter and window overlap. An alternative strategy is to have one window containing a small overview, while a second window shows a large more detailed view [3, 10]. The small overview contains a rectangle that can be moved and resized, and its contents are shown at a larger scale in the large view. This strategy, however, requires extra space for the overview and forces the viewer to mentally integrate the detail and context views. An operational overhead is also required, because the user must regularly move the mouse between the detail and context windows. Speed-dependent automatic zooming (SDAZ) is a navigation technique first pro- posed by Igarashi & Hinckley [14]. It couples rate-based scrolling with automatic zooming to overcome the limitations of typical scrolling interfaces and to prevent ex- treme visual flow. This means that as a user scrolls faster the system automatically zooms out, providing a constant information flow across the screen. This allows users to efficiently scroll a document without having to manually switch between zooming and scrolling or becoming disoriented by fast visual flow, and results in a smooth curve in the space-scale diagram. In traditional manual zooming interfaces, the user has to interleave zooming and scrolling (or panning); thus the resulting pan-zoom tra- jectory forms a zigzag line. Cockburn et al. [7, 8, 22, 25, 26] presented further devel- opments, with a usability study of performance-improved SDAZ prototypes. 3 Dynamics and interaction In this paper we use systems of differential equations to describe the interaction be- tween user and computer. Skeptics might question this “Why introduce dynamics, when dynamic systems tend to be more difficult to control than static ones? Vehicle control systems tend to go to great trouble to hide the underlying dynamics of the ve- hicle from the driver.” We explicitly include dynamics because we can only control what we can perceive, and while, in principle, we can navigate instantly in an arbitrary information space, given a static interaction mechanism (e.g. clicking on a scroll bar), if we are depend- ent on feedback to be displayed while pursuing our goals, there will be upper limits on the speed at which the display can change. This is especially true in cases where there is uncertainty in the user’s mind about where to go, and when they have the op- tion to change their goal on route, as more information becomes available. In order to cope with this, interface designers have a long history of hand-crafting transition ef- fects in a case-by-case manner. Nonlinear mouse transfer functions are long- established examples of finely-tuned dynamic systems driven by user input. One of our long-term goals is to investigate whether describing the dynamics of in- teraction using the tools of control engineers allows us a more consistent approach to analyzing, developing and comparing the ‘look-and-feel’ of an interface, or in control terms, the ‘handling qualities’ . Control synthesis often focuses on analysis of cou- pling among system states. Speed-dependent zooming is an obvious example of this, but if we generalize the approach to other interaction scenarios, with possibly a larger number of interacting states/inputs, we will require more general methods to analyse the consequences of coupling effects. Control methods are likely to be especially im- portant for design for mobile devices, where sensor noise, disturbance rejection, sen- sor fusion, adaptive self-calibration and incorporating models of human control be- haviour are all important research challenges. In cases such as the use of accelerometers as input devices, the direct mapping of acceleration in the real world to acceleration in the interface provides an intuitive mapping, which also suggests a range of other affordances, especially for multi-modal feedback, which can then be utilized by interface designers. Real-world effects such as haptic feedback of springs, or friction linked to speed of motion are easy to repro- duce in a dynamic system, and we can choose to explicitly use these features to de- sign the system to encourage interaction to fall into a comfortable, natural rhythm.